* In the specification, you refer to "t_start". I guess you mean "time_start"?

* Miners can, especially when close to a block doubling or shortly
after activation, to some extent manipulate max block size by
manipulating the time stamp in the block header within valid limits.
According to the pseudo code in the specification, the first and a
handful of subsequent blocks after activation could actually have
negative max block sizes due to this (depending on how you define the
% operator of the pseudo code). I haven't checked the reference
implementation, but I do think that the specification section should
explicitly handle this.